Session: Session 02-05: Two Phase Cooling - II

Paper Number: 173212

173212 - Integrated Energy-Thermal Simulation and Two-Phase Flow Network Modeling for Data Center Cooling 

The increasing demand for AI, cloud computing, and high-performance computing has elevated data centers to critical infrastructure status. This surge in computational intensity is driving the transition to higher rack power densities, creating major challenges for thermal management. In many data centers, cooling now accounts for 40 to 50 percent of total energy consumption. There is a growing need for simulation tools that can evaluate cooling strategies not only from a thermal perspective but also in terms of their energy impact, particularly during early-stage design when key architectural decisions are made.

This work presents recent progress under the MOSTCOOL project, which aims to build a modular simulation environment for analyzing and optimizing data center cooling systems. The current implementation integrates reduced-order thermal models with EnergyPlus to simulate air-cooled data centers. Through this integration, users can evaluate the impact of changing CRAH supply temperatures and economizer thresholds on both temperature distributions and facility-scale energy use. The reduced-order thermal model predicts spatial temperature fields in the data center based on rack-level heat loads, while the coupled EnergyPlus simulation computes cooling energy consumption under varying weather conditions and system configurations.

Alongside this integrated air-side simulation capability, a new modeling tool is under development for two-phase liquid cooling systems. This tool focuses on system-level modeling of liquid cooling loops using a network representation. Each cooling loop is modeled as a series of connected components such as heat sinks, heat exchangers, pipes, and pumps. The solver calculates steady-state pressure, mass flow rate, and enthalpy across the network by applying conservation equations for mass, momentum, and energy. The tool is already capable of handling flow splits and merges, and has been applied to test cases involving non-uniform heat loads across branches.

While the two-phase model is currently developed as a standalone tool, the long-term plan is to integrate it with the existing co-simulation platform to allow simultaneous evaluation of air and liquid cooling systems in a unified simulation environment. This would support hybrid cooling architectures where air cooling is used for lightly loaded zones and liquid cooling is applied to high-density racks.

The presentation will cover two primary components. The first is the current implementation of the air-cooling co-simulation platform, including its structure and results from case studies that investigate how cooling energy use changes with supply temperature and economizer operation. The second is the development and early results of the two-phase flow network model, including examples with branched topologies and non-uniform thermal inputs that reflect realistic data center layouts.

Together, these efforts aim to provide a scalable, scientifically grounded framework for evaluating data center cooling strategies under realistic operating conditions. The current integration of thermal and energy simulation is already yielding actionable insights for air-cooled systems. The ongoing development of the two-phase modeling capability extends this approach to support emerging cooling technologies that are essential for next-generation, high-density data centers.

Presenting Author: Roshith Mittakolu University of Maryland, College Park

Presenting Author Biography: Roshith Mittakolu is a doctoral researcher in Mechanical Engineering at the University of Maryland, specializing in thermal-fluid sciences and electronics cooling. His work focuses on developing high-fidelity and reduced-order models for data center thermal management, with an emphasis on co-simulation, two-phase flow network modeling, and digital twin integration. Roshith is a core contributor to the MOSTCOOL project, a multi-institutional initiative aimed at improving energy efficiency, reliability, and cost modeling in next-generation data centers. His research blends computational modeling, systems integration, and software development to enable more intelligent design and operation of cooling infrastructure.